Dataset Card for xMINDlarge

Dataset Summary

xMINDlarge is an open, large-scale multi-parallel news dataset for multi- and cross-lingual news recommendation. It is derived from the English MINDlarge dataset using open-source neural machine translation (i.e., NLLB 3.3B).

For the small version of the dataset, see xMINDsmall.

Uses

This dataset can be used for machine translation, text retrieval, or as a benchmark dataset for news recommendation.

Languages

xMIND contains news translated into 14 linguistically and geographically diverse languages, with digital footprints of varying sizes.

Code	Language	Script	Macro-area	Family	Genus
FIN	Finnish	Latin	Eurasia	Uralic	Finnic
GRN	Guarani	Latin	South-America	Tupian	Maweti-Guarani
HAT	Haitian Creole	Latin	North-America	Indo-European	Creoles and Pidgins
IND	Indonesian	Latin	Papunesia	Austronesian	Malayo-Sumbawan
JPN	Japanese	Japanese	Eurasia	Japonic	Japanesic
KAT	Georgian	Georgian	Eurasia	Kartvelic	Georgian-Zan
RON	Romanian	Latin	Eurasia	Indo-European	Romance
SOM	Somali	Latin	Africa	Afro-Asiatic	Lowland East Cushitic
SWH	Swahili	Latin	Africa	Niger-Congo	Bantu
TAM	Tamil	Tamil	Eurasia	Dravidian	Dravidian
THA	Thai	Thai	Eurasia	Tai-Kadai	Kam-Tai
TUR	Turkish	Latin	Eurasia	Altaic	Turkic
VIE	Vietnamese	Latin	Eurasia	Austro-Asiatic	Vietic
ZHO	Mandarin Chinese	Han	Eurasia	Sino-Tibetan	Sinitic

Dataset Structure

Data Instances

>>> from datasets import load_dataset
>>> data = load_dataset('aiana94/xMINDlarge', 'ron')

# Please, specify the language code.

# A data point example is below:

{
"nid": "N49265"
"title": "Aceste reţete cu sos de afine sunt perfecte pentru cina de Ziua Recunoştinţei.",
"abstract": "Nu vei mai vrea niciodată versiunea cumpărată din magazin."
}

Data Fields

• nid (string): news ID (same as in the MIND dataset)
• title (string): news title
• abstract (string) : news abstract (optional)

Data Splits

For all languages, there are three split: train, dev, test.

Dataset Creation

Source Data

The news were machine-translated from the MINDlarge dataset.

Data Collection and Processing

We translated the news articles using the open-source model NLLB 3.3B. For more details regarding the translation setup and data quality, we refer to the corresponding paper.

Personal and Sensitive Information

The data is sourced from newspaper sources and contains mentions of public figures and individuals.

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

Users should keep in mind that the dataset contains short news texts (e.g., news titles and abstracts), which might limit the applicability of the developed systems to other domains.

Additional Information

Licensing Information

The dataset is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. If you intend to use, adapt, or share xMINDlarge, particularly together with additional news and click behavior information from the original MIND dataset, please read and reference the Microsoft Research License Terms of MIND.

Citation Infomation

BibTeX:

@inproceedings{iana2024mind,
  title={Mind your language: a multilingual dataset for cross-lingual news recommendation},
  author={Iana, Andreea and Glava{\v{s}}, Goran and Paulheim, Heiko},
  booktitle={Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval},
  pages={553--563},
  year={2024}
}

Also consider citing the following:

@inproceedings{wu2020mind,
  title={Mind: A large-scale dataset for news recommendation},
  author={Wu, Fangzhao and Qiao, Ying and Chen, Jiun-Hung and Wu, Chuhan and Qi, Tao and Lian, Jianxun and Liu, Danyang and Xie, Xing and Gao, Jianfeng and Wu, Winnie and others},
  booktitle={Proceedings of the 58th annual meeting of the association for computational linguistics},
  pages={3597--3606},
  year={2020}
}